Lin, H. C. K., Su, S. H., Chao, C. J., Hsieh, C. Y., & Tsai, S. C. (2016). Construction of Multi-mode Affective Learning System: Taking Affective Design as an Example. Educational Technology & Society, 19 (2), 132–147. Construction of Multi-mode Affective Learning System: Taking Affective Design as an Example Hao-Chiang Koong Lin1, Sheng-Hsiung Su1*, Ching-Ju Chao2, Cheng-Yen Hsieh1 and Shang-Chin Tsai1 1 Department of Information and Learning Technology, National University of Tainan, Tainan, Taiwan // 2 Department of Applied Foreign Languages, Tung Fang Design University, Kaohsiung, Taiwan // koong000@ms39.hinet.net // ice.shsu@gmail.com // chingju@mail.tf.edu.tw // allen01816@hotmail.com // maney100@hotmail.com * Corresponding author ABSTRACT This study aims to design a non-simultaneous distance instruction system with affective computing, which integrates interactive agent technology with the curricular instruction of affective design. The research subjects were 78 students, and prototype assessment and final assessment were adopted to assess the interface and usability of the system. Prototype assessment consisted of heuristic assessment and the system usability scale, while final assessment adopted the triangular cross-validation method: where the questionnaire for user interaction satisfaction, observation, and interviews were used to explore the effect of learning and obtain qualitative and quantitative information for analysis. According to the experimental results, the usability of the non-simultaneous distance instruction system with affective computing was high; the respondents showed highlevel satisfaction regarding interaction with the affective learning system; the training game response mechanism of the system could effectively improve the emotion of learning; there was a significant improvement in the effect of learning based on the affective learning system. Keywords Affective computing, Affective tutoring system, Text emotion, Facial expression, Skin potential Introduction In recent years, the prevalence of smart devices demonstrates a lifestyle associated with computers in modern society. Regarding different types of program, besides functions, humanity is also critical. The most significant difference between human beings, computers, and machines is that human beings have emotion. Therefore, in order to reinforce the humanity of computers and machines, they should approach human emotion in order to provide proper feedback. With the popularity of the internet, non-simultaneous distant instruction gradually replaces traditional instruction. Emotion is part of the key semantic information of interpersonal relationships. Positive emotion leads to more successful learning processes (Ezhilarasi & Minu, 2012; Eyharabide & Amandi, 2012). Instructional systems can automatically recognize a learners’ emotion and provide proper feedback, thus, it is a significantly potential and essential research indicator (Islam, 2013). With the prevalence of online learning, if computers can recognize learners’ learning emotion and maintain their position emotion, it will result in learning efficiency and effectiveness (Kerr, Rynearson, & Kerr, 2006). In interpersonal interaction, conversation and facial expression are the most direct methods to recognize others’ emotions; however, people can disguise these two emotions. Therefore, aside from introducing semantic and facial expression emotional recognition, this study measures learners’ physical signals as indicators of emotional judgments. Since physical signals are non-volitional physical reactions, the emotional state is the most objective. The introduction of multi-mode Affective Computing can enhance the precision rate of emotional recognition in Affective Tutoring Systems. This study applies Affective Computing, as developed by the laboratory, to research the findings and development of Intelligent Tutoring Systems. Through developed Affective Tutoring Systems, it expands different dimensions and functions. The purpose is to strengthen learners’ trust and involvement in learning by a more advanced and complete emotion recognition module. By complete and effective assessment, it demonstrates the value and existence of this study. ISSN 1436-4522 (online) and 1176-3647 (print). This article of the Journal of Educational Technology & Society is available under Creative Commons CC-BY-ND-NC 3.0 license (https://creativecommons.org/licenses/by-nc-nd/3.0/). For further queries, please contact Journal Editors at ets-editors@ifets.info. 132 Literature review Affective tutoring systems Changes of technology are rapid and important, and various kinds of digital technologies and techniques are gradually introduced to instructional environments. As the traditional teachers’ one-to-many instruction model changes, learning models have become diverse (Chen, Kao, & Sheu, 2003). ITS (Intelligent Tutoring Systems) means to provide personalized instruction by computer analysis or direct feedback to students. ITS finishes different instructional tasks by simulating teachers. According to different students’ characteristics and states, it indicates various instructional methods. Affective Computing is a research field that deals with issues of emotions and computers. Generally speaking, it is classified into four research levels: emotional recognition, emotional expression, having emotion, and emotional intelligence. Most studies focused on emotional recognition (Picard & Klein, 2002). Figure 1. Framework of affective tutoring (Source: Ammar et al., 2010) Technology introduction reinforces class activities. In addition, teachers can observe the negotiation, communication, cooperation, and interaction among students (Liaw, Chen & Huang, 2008; Infante et al., 2009; Martin, Pastore, & Snider, 2012). However, without an appropriate learning strategy, learning effectiveness will not be as expected (Peng et al., 2009). ATS (Affective Tutoring Systems) means the defection of students’ learning and emotional states to offer proper emotional feedback and regulate students’ learning emotion (Mao & Li, 2010; Hsu, Lin, Lin, & Lin, 2014). ATS is developed upon ITS, and aims to effectively adapt to students’ emotion by simulating human beings (Ammar, Neji, Alimi, & Gouardères, 2010; Lin, Wang, Chao, & Chien, 2012). Although ATS is developed in recent years, it is the first research that adapts to and recognizes emotion. Thus, this study reviewed this study of Picard (Picard, 2000), who proposed a conceptual module that influences learning emotion, and constructed a system that recognizes learners’ emotional state, provides appropriate feedback, and reinforces learners’ learning (Lin, Chen, Sun, & Tsai, 2012; Lin, Hsieh, Loh & Wang, 2012). With the Affective Computing technique, computers can recognize human emotions. ATS is considered as a personalized training model. Lin, Tsai, Cheng, Chao, and Su (2014) combined Affective Computing with a webpage system to develop Affective Computing on a webpage to provide adaptive learning for learners. Mao and Li (2010) undertook an investigation into the key factors that influenced students’ satisfaction when using the affective learning system, and found that the factors included the attitude of students, the effect of tutoring, the accuracy of emotion recognition, the quantity of identifiable emotions, instructive action, and the usability of the system. Sarrafzadeh, Alexander, Dadgostar, Fan, and Bigdeli (2008) developed a system targeting math for primary school students. Through the instruction of a lifelike animated agent, the system analyzed the facial expressions of students to identify their emotions, and showed the emotion of the animated agent. By incorporating affective computing into 133 a smart tutoring system, Ammar et al. (2010) detected and judged facial expressions and computed emotions, in order to enhance the interaction between instruction and learners, develop students’ interest in learning, help them absorb knowledge, and significantly increase the effect of learning through mutual assistance among learners. The framework chart of the affective learning system is as shown in Figure 1. Emotional recognition Chinese semantic information Emotion is the key semantic information of interpersonal interaction (Ezhilarasi & Minu, 2012). If emotion can be precisely recognized, it will help to make decisions in better way (Lin, Wu, & Hsueh, 2014). The first condition of text emotion recognition is to understand semantic content to acquire precise information. It should be based on natural language processing and semantic analysis (Yan, Bracewell, Ren, & Kuroiwa, 2008). The assumption of document frequency might lower the precise rate of classification, as many terms with high document frequency are usually unused words or unimportant information (Basu & Murthy, 2012). Therefore, when selecting terms by document frequency, other methods are usually adopted. Lu, Lin, Liu, Cruz-Lara, and Hong (2010) proposed the automatic and hierarchical emotional semantic acquisition system, which is highly intensive. Through an independent database, it automatically recognizes the subjects’ semantic emotional responses to the events of sentences. Regarding classification of emotions, there are two principle methods: monitoring and non-monitoring learning (Feldman, 2013). Studies on emotional analysis demonstrated that the importance of the terms, as calculated by TF-IDF weights, is highly effective (Liu, 2012). Facial expression Facial recognition techniques are commonly applied in daily lives (Gunes & Piccardi, 2007), such as, automatic facial focus, domestic security and recognition, and facial recognition guard systems, which unlock the door of facial recognition. Camera shutters with multi-smiles is another extended technique of facial recognition. Therefore, it demonstrates the common application of facial expressions and potential. According to faces and facial features, there are generally six kinds of facial expressions: joy, anger, sadness, surprise, fear, and disgust (Ekman & Friesen, 1971; Ezhilarasi & Minu, 2012). Metri, Ghropade, and Butalia (2012) constructed a facial emotional recognition system by facial features, as proposed by Ekman, and enhanced the effect of recognition of emotions through physical poses. Figure 2 shows the steps of recognizing facial expressions. Hsu et al. (2014) integrated facial recognition and semantic recognition by multi-mode for Affective Computing to develop an affective tutoring system. Figure 2. Steps of facial emotional recognition (Source: Metri, Ghorpade, & Butalia, 2012) Skin conductance degree Skin conductance is also called Galvanic Skin Response (GSR), and it means the electronic conductance on skin. When people’s emotions change, their body actions, facial expressions, and physical reactions will change accordingly. For instance, with stimulus, the secretion of skin sweat glands will influence changes of the Galvanic Skin Response. We can measure the change of skin potential by Affectiva Q sensor. Through this instrument, we can obtain the figures and analyze the different waves. The peak of the wave means severely positive and negative emotional reactions, as shown in Figure 3. 134 Figure 3. Galvanic skin response Research method Research process and implementation steps In order to probe into learning effectiveness and system usability of the multi-mode Affective Tutoring System, as introduced in digital art materials, which include network art, dynamic video-audio technology, recording art, software art, and new media art, 78 college students of a national university in southern Taiwan were invited as the research subjects. Their educational levels were college and university to master. By prototype assessment and final assessment, it evaluated the system interface and system usability. Figure 4 shows the system assessment process of this study. The first step is to put forward a system concept model for the design of the prototype system. After the design, the prototype system is assessed. The assessment of the prototype system consists of two parts--the usability scale of the assessment system for average users, and the assessment of experts. The last step is the final assessment, which adopts the triangular cross-validation method. The research analysis is based on questionnaires, observations, and interviews. Figure 4. Process of system assessment Prototype system assessment • Assessment of average users--System usability scale: To obtain information regarding the subjective feeling of average users during operation of the system, this study adopted the system usability scale (SUS) as an assessment tool. Developed by the British Digital Equipment Co., Ltd. in 1986, the scale was designed to inform enterprises of the general usability of their products and provide a low-cost, reliable, and fast method. It is a 135 • • five-level Likert scale, with each item including five options ranked in an ascending manner: (1) Strongly disagree; (2) Disagree; (3) Average; (4) Agree; (5) Strongly agree. The scale comprises 10 items and adopts the forward and backward cross-questioning method. Usually, respondents complete the scale without discussion after operating the system. Through a formula, scoring will be converted into one with “100 points” as the full mark, where a higher score indicates stronger satisfaction of respondents for the system. Assessment of experts--Heuristic assessment: Heuristic assessment is an informal usability testing method used to detect problems regarding usability in the design of a user interface, in order that these problems can be regarded as a focused part of the redesign. In the heuristic assessment, experts follow a group of given usability heuristics, and evaluate the constituents of a respondent interface to see if these constituents are consistent with these heuristics (Nielsen, 1994). Meanwhile, Nielsen suggested inviting 3-5 evaluators, and predicted about 75% of the problems regarding usability. Final assessment—Triangulation: As a method of testing research information, triangulation adopts more than two resources to obtain full understanding, and demonstrate a specific reference point or topic, with the aim of enhancing the rigorousness and reliability of research. In general, it is recommendable to include three information sources, which allows an evaluation with diverse perspectives, and provides a neutral stance when two views are contradictory. Affective tutoring systems The system framework of this study is as shown in Figure 5: Figure 5. System framework Emotional recognition module • Semantic recognition module: By dialogue between the subjects and the interactive agent, this study conducts Chinese semantic emotional recognition; and through the dialogue input by the subjects, the subjects’ emotional state is immediately recognized. The construction is as shown below: (1) construction of emotional dictionary; (2) semantic structure message processing: word segmentation rules, semantic structure, message processing, keywords matching, and word segmentation; (3) acquisition of semantic emotion. Figure 6 shows the process of semantic analysis. 136 Figure 6. Process of semantic analysis • • Facial expression recognition module: By open library--EmguCV, this study develops a facial expression recognition module. EmguCV is an OpenCV component packaged by C# for the development of a Visual Studio. EmguCV not only has powerful image processing capacity, but is also an open and free library, thus, system development is less difficult. The steps of facial expression recognition are shown as follows: (1) recognition of human beings’ facial positions; (2) recognized facial features are compared to six kinds of facial expressions by classification - HaarTraining and classification; (3) recognized facial expressions refer to emotions; in the opposition situation, there is no emotion. Physiological signal module: By Q sensor, this study includes the skin conductance information of the physical signals in the system. Through a Bluetooth connection, Q sensor sets the device as a COM Port. By Visual Studio C#, it obtains related figures. Figure 7 shows the process of physiological signal. Figure 7. Process of physiological signal • Brainwave concentration and relaxation training module: By NeuroSky MindWave Mobile, this study conducts brainwave concentration and relaxation training. When the subjects learn by this system, they rely on the emotional recognition of the previous three kinds of emotional recognition modules. When the subjects have negative emotion, the system will automatically accumulate the emotion. When a negative emotion unexpectedly occurs, the system will automatically stop the teaching material and record the learning time in the 137 • database. It then activates the Brainwave Visualizer, as developed by NeuroSky, for the subjects’ concentration and relaxation training. With training effectiveness, the subjects can continue learning. System interface: In this study, the system interface design is divided into 7 zones: function, teaching material, interactive agent, semantic dialogue, facial expression, skin conductance signal, and system record. Figure 8 shows the layout of the system interface. Figure 8. Layout of system interface (a) Tools (see Figure 8 a): video control, which can be played, paused, and stopped; after-class questionnaires include learning effectiveness, scale of system usability, and user interaction satisfaction questionnaires. After learning the subjects, there can be after-class assessment and system assessment; regarding parameter settings, the subjects can regulate facial expression recognition parameters. According to individual conditions, it regulates recognition sensitivity. There are six kinds of recognition, and each kind includes two regulation parameters; the description column provides system instructions, system history, and laboratory introduction. The end button of this system is termination of the course. Three questionnaires will be completed to finish the experiment. (b) Teaching material (see Figure 8 b): the video playing of teaching material. The videos, as recorded by a teacher in advance, are constructed for the subjects’ learning. Video of this experiment is based on system design. Therefore, the affective design is treated as the instructional content. A combination of instruction content and system allows the subjects to have profound learning experiences. (c) System record (see Figure 8 c): records all activities of this system, including system starting time, change of course playing, database connection records, results of facial expression recognition, semantic and emotional recognition, Q sensor connection, and accumulation of negative emotion. The recording format is “hour: minute: second. Milli-second recording”. (d) Facial expression (see Figure 8 d): the subjects’ facial expressions are captured by webcam. When the face and facial expression are captured, a square frame will be drawn. In the figure, the white frame is the subjects’ captured face; the yellow frame is the subjects’ facial expression recognized as joy. (e) Skin conductance (see Figure 8 e): the subjects’ skin conductance is captured by Q sensor. Captured frequency is 32 times/every second. The system shows the total in the system and conducts emotional sensing. (f) Interactive agent (see Figure 8 f): the agent designed by this study: batman. According to the emotion recognized by this system, it transfers the movement of facial expressions and interacts with the subjects in a semantic dialogue zone. 138 (g) Semantic dialogue (see Figure 8 g): the subjects can interact with the interactive agent in this zone. The system will conduct semantic recognition according to sentences input by subjects, and provide immediate feedback. (h) Researcher’s operation (see Figure 8 h): records the subjects’ time of system learning and current time. Two buttons are function module controls. The researcher assists with the operation button. • Learning process: First of all, interactive agent module: this module is the bridge between a learning system and the subjects. By an agent mechanism, the system can interact with the subjects and properly provide feedback. Second, video course teaching material module: this module is the video course teaching material recorded by a teacher in advance. It is based on video, and the subjects learn online, as in a classroom. Third, learning state recording module: this module is the core of this system. The system automatically and completely saves the subjects’ learning stages and uploads it to a database. Thus, the teacher can immediately recognize each subject’s learning condition, and actively solve problems for the subjects after class in order to reinforce learning effectiveness. Experimental results and analysis To determine if this non-simultaneous distance instruction system with affective computing could improve the effect of learning, and evaluate the usability of the system, this study invited 78 undergraduates and graduates to participate in the experiment. The respondents were divided into two groups--the Experimental group, which used the affective learning system, and the Control group, which adopted the online webpage learning system. The course materials were videos pre-recorded by the teachers of a university in Kaohsiung, Taiwan. Figure 9 shows the flow chart of the experiment. Prior to learning, the respondents received a learning effectiveness assessment (pre-learning test), and then watched a 16 minute instructive video. After learning, the respondents received another learning effectiveness assessment (post-learning test), completed the questionnaire for user interaction satisfaction, and took the interview. The entire experiment was recorded, and final assessment--triangulation was conducted after the experiment. Figure 9. Flow chart of the experiment 139 The system counted the negative emotions captured in the learning of the Experimental group, and the information was added into the learning status database. When the negative emotion reaches a certain level, the system will suspend the course and start a brainwave concentration and relaxation training game. In concentration training, stronger concentration will result in the explosion of a bucket; while sustainable and strong concentration will maintain the explosion of the bucket, and the best explosion time will appear on the interface, as shown in Figure 10. In relaxation training, if the respondents feel more relaxed, a balloon will start to rise; if the respondents maintain a high-level of relaxation, the balloon will float above and rise higher, and the maximum height the balloon reaches will appear on the interface, as shown in Figure 11 The respondents can decide if they want to continue brainwave concentration and relaxation training; if they refuse to continue the training, they can shift the interface to the affective learning system to continue learning. Figure 10. Brainwave concentration training Figure 11. Brainwave relaxation training 140 After learning, both the Experimental group and the Control group immediately received the learning effectiveness assessment (post-learning test), and then completed the questionnaire for user interaction satisfaction. After completing the questionnaire, the Control group finished its task in the experiment, while the Experimental group finished its task only after the interview. System usability analysis At the prototype system development phase, this study invites 30 users for prototype system usability analysis. Cronbach’s α acquired by the scale of system usability is .791, and the least reliability of .7 is accepted by this study, as it demonstrates that the reliability of the questionnaires is acceptable. After transforming reverse responses into positive responses in the scale of system usability, this study conducts item analysis, as shown in Table 1. Noticeably, regarding Q4 and Q10, only 46.6% and 66.8% users, respectively, suggest that they do not need assistance, and the subjects indicate that they can use the system without prior knowledge. Hence, the system interface must be simplified for the subjects’ ease of use. The researcher analyzed the questionnaire filled by the respondents and converted the backward questions in the system usability scale into forward ones for analysis. According to the 5-point scale, the researcher selected the two highest score -- the percentages of “4” and “5” for aggregation analysis. As is shown in the following table, 56.7% of the respondents to Q1 were willing to use the affective learning system on a regular basis; 70.2% of the respondents to Q2 did not believe that the system was too complicated; 83.3% of the respondents to Q3 thought that the system was easy to use; 46.6% of the respondents to Q4 believed that the system entailed little assistance from technicians; 73.4% of the respondents to Q5 thought that all the functions of the system were well integrated; 83.3% of the respondents to Q6 did not believe that the system was inconsistent; 86.6% of the respondents to Q7 thought that most people would master the skills of using the system within a short time; 96.7% of the respondents to Q8 did not believe that the system was too difficult to use; 96.7% of the respondents to Q9 were confident that they could use the system; 66.8% of the respondents to Q10 did not think it necessary to acquire much knowledge to use the system. Items Mean Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 Q10 AVERAGE 3.73 3.80 4.13 3.20 3.87 4.10 4.20 4.40 4.27 3.80 3.95 Table 1. Analysis of system usability scale 5-point scale percentages (%) Standard deviation 1 2 3 4 .740 0 0 43.3 40.0 .925 3.3 3.3 23.2 50.0 .681 0 0 16.7 53.3 1.064 6.6 20.2 26.6 40.0 .629 0 0 26.6 60.0 .759 0 3.3 13.4 53.3 .664 0 0 13.4 53.3 .563 0 0 3.3 53.3 .521 0 0 3.3 66.7 .847 0 6.6 26.6 46.6 .7393 .99 3.34 19.64 51.65 5 16.7 20.2 30.0 6.6 13.4 30.0 33.3 43.4 30.0 20.2 24.38 The scores in the scale were obtained according to the scoring of the system usability scale. They reflected the comprehensive assessment of the respondents on the system usability and can be used for the comparison of usability among different versions of the system. The total score (ranging from 0 to 100 points) of the questionnaire was obtained according to the scoring of the system usability: (1) the score of the items labeled with odd numbers was obtained by subtracted 1 from the original score; (2) the score of the items labeled with even numbers was obtained by subtracted 5 from the original score; (3) the total score (ranging from 0 to 100 points) of the questionnaire was obtained by first aggregating the scores of all items and then multiplying the aggregated score with 2.5. Based on the scoring of the system usability scale, the researcher conducted statistics of questionnaire scores of the subjects. The result is as shown in Table 2; where the mean is 73.75, the median is 70.00, the mode is 67.50, the standard deviation is 11.14, and the minimum and maximum are 47.50 and 95 respectively. The score for the system usability is 70, which indicates that most of the respondents were satisfied with the system usability. The average score for the system is “73.75 points,” which shows that average respondents were satisfied with the usability of system. Meanwhile, it was compared with Figure 12, which fell into the zones of “good” and “excellent,” respectively. 141 Number of samples 30 Table 2. Conversion results of scores in the system usability scale Mean Median Mode Standard deviation Minimum 73.75 70.00 67.50 11.14 47.50 Maximum 95.00 Figure 12. Score distribution of the system usability (Source: Bangor, Kortum, & Miller, 2009) User interaction satisfaction analysis The questionnaire for user interaction satisfaction includes 6 dimensions: total use reaction, display of screen, terms and system information, learning, system function and usability, and user interface. Each dimension includes 2~6 items. According to the subjects’ satisfaction, the rating is from 1~7. There are a total of 27 items. Cronbach’s α of the user interaction satisfaction questionnaire is .949, and the least reliability of .9 accepted by this study. Thus, the reliability of questionnaires results is good. This study analyzes questionnaires responded by the subjects, and obtains the mean by the total scores of the items of the dimensions divided by the number of items. It then conducts descriptive statistical analysis on the results, with the analytical outcome as shown in Table 3. The means of the dimensions of the Experimental group are higher than those of the Control group. Satisfaction with dimensions is at least 5 in the 7-point scale, meaning that the subjects’ subjective satisfaction with the design of the human-machine interface of the system is good. Regarding mean, we compare satisfaction with the dimensions of the Experimental and Control groups. Satisfaction with using Affective Tutoring Systems is higher than online webpage learning systems, which shows that, in the same video material learning, satisfaction with Affective Computing recognition is higher than non-simultaneous distant instructional systems. In addition, regarding the means of total use reaction, usability, and user interface, the Experimental group is more significant than the Control group, which shows that users’ interaction satisfaction with the system is high. Table 3. User interaction satisfaction questionnaire--descriptive statistics Experimental group (30 people) Control group (48 people) Dimensions of questionnaires Mean Standard deviation Mean Standard deviation Total usage reaction 5.14 .83 4.40 .75 Display of screen 5.76 .92 5.09 .89 Term and system information 5.29 .69 4.78 .99 Learning 5.49 .83 5.00 .88 System function 5.37 .99 5.02 1.08 Usability and user interface 5.38 .89 4.81 .89 Average 5.405 .73 4.852 .74 According to questionnaire results and t testing of independent samples, this study attempts to determine if satisfaction is different between the Experimental group and the Control group. Analytical results are as shown in Table 4. Assessment results of user interaction satisfaction is Experimental group > Control group and significance .002 < .05, meaning that the satisfaction of the two groups’ is significantly different. Based on previous results, the Experimental and Control groups have significantly different satisfaction using the Affective Tutoring 142 System and online webpage learning system. Interaction satisfaction with Affective Tutoring Systems is higher than online webpage learning systems. Table 4. User interaction satisfaction questionnaires—t-test of independent samples Number of Standard Mean t value Group samples deviation Experimental group 30 5.405 .73 -3.238 Control group 48 4.852 .74 Note. **p < .01. Significance (two-tailed) .002** Analysis of learning effectiveness Learning effectiveness before the experiment To determine the difference in knowledge between the Experimental group and the Control group, the researcher undertook independent sample t testing on the pre-learning scores of the respondents. As shown in Table 5, significance (.38 > .05) indicates that there was no significant difference between the two groups before learning, meaning that both groups shared similar knowledge before the experiment. Table 5. One-way ANOVA of learning effectiveness before the experiment Number of Standard Mean t value Group samples deviation Experimental group 30 47.00 16.43 .878 Control group 48 50.21 15.23 Significance (two-tailed) .383 Learning effectiveness after the experiment The researcher undertook one-way ANOVA of the post-learning scores of the respondents to determine if there was any significant difference between the two groups, and the result is as shown in Tables 6 and 7. The average postlearning scores of the Experimental group and the Control group were 73.00 and 63.13, respectively, which is higher than the pre-learning average score of 47.00 and 50.21, respectively. The significance (.019 < .05) of the postlearning score manifests that the instruction played a significant role in the improvement of learning effectiveness. Group Experimental group Control group Total Source of variance Inter-group In the group Total Note. *p < .05. Table 6. Descriptive statistics of the post-experiment learning effectiveness Number of samples Mean Standard deviation 30 73.00 13.17 48 63.13 20.02 78 66.92 18.26 Table 7. One-way ANOVA of the post-experiment learning effectiveness Quadratic sum Freedom Mean quadratic sum F 1800.288 1 1800.288 5.734 23861.250 76 313.964 25661.538 77 Significance .019* The researcher subtracted the pre-learning score from the post-learning score, and conducted percentage conversion for independent sample t testing, and the result is as shown in Table 8. The mean of the increase in the learning effectiveness of the Experimental group is 94.94%, whereas, that of the Control group is 29.60%. The significance of (.033 < .05) means that there was significant increase in the learning effectiveness of the two groups, which indicates that the increase in the learning effectiveness of the Experimental group was greater than that of the Control group, and the respondents could improve their learning effectiveness with the affective learning system. 143 Group Experimental group Control group Note. *p < .05. Table 8. Comparison of the increase in learning effectiveness Number of Standard Standard Mean t value samples deviation error 30 94.94 158.09 28.86 -2.228 48 29.60 35.98 5.19 Significance (two-tailed) .033* Results of the interview After the experiment, the respondents took an interview that lasted for 3 to 5 minutes. The researcher wrote the respondents’ answers to the above interview questions, which have the following results: • Most respondents would first turn to their partners for help, and then their teachers, when encountering problems in learning. The majority of the respondents thought it was difficult to communicate with teachers, and thus, would not regard teachers as the first choice when they sought help. Only a few respondents did not turn to teachers or partners for help; instead, they sought help on the Internet. • Most respondents thought that active help from teachers would facilitate their learning; however, only a limited number of teachers can notice the problems of students and voluntarily offer help in the current educational environment. A few respondents believed that they would not seek help until they face problems, and that active help from teachers might upset or embarrass them. • Most respondents thought it necessary to keep the mechanism of Question (2). According to them, not all students would actively seek help, thus, the mechanism was good. A handful of students still preferred to seek help on the Internet, as they thought such a method could improve their learning and memory. • Most respondents thought that the response from the interactive agent was too limited, and that an interactive agent would give the same response to similar emotion. • Most respondents believed that the brainwave training game was interesting; however, some thought the game was so difficult that they could not fulfill the objectives, and thus, felt frustrated; while others were deeply fascinated by the training, and hoped to break the record kept by themselves or others. • Most respondents thought that what was taught was boring, and such a mechanism could help them refocus on what was to be taught later. Only a few respondents believed that the game interrupted their learning, was of little help in their learning, and they felt reluctant to continue learning once they started to play the game. According to the results of the interviews, most respondents were satisfied with the affective learning system, and thought that the learning mechanism, as proposed by the researcher, was effective and necessary. In emotion recognition, the number of keywords for emotion recognition of Chinese meaning was too small, and the system failed to give responses to many daily expressions; the recognition of facial expression was too sensitive; the information about facial expressions used for classification was inadequate, and errors were likely to occur when the system judged a face for the first time. One respondent thought that the interactive agent distracted his attention during learning, and that the brainwave game would interrupt learning. Conclusion and future studies This study incorporated affective computing into a non-simultaneous distance instruction system, and adopted the multi-mode affective perception module to detect the emotions of the respondents during learning. The curricular collocation system design used affective design as the learning material to provide respondents with more impressive learning experiences. Prototype assessment and final assessment were employed to discuss the usability of the system, as well as satisfaction with the system. The former involved heuristic assessment based on expert evaluation, where a system usability scale was used to analyze the usability of the revised system; the latter adopted triangulation, where methods such as questionnaires for user interaction satisfaction, and observations, as well as interviews, were used to explore satisfaction with the system; the researcher designed the assessment of learning effectiveness according to the learning material in order to analyze the learning effectiveness of the respondents. Based on the experiment results, the researcher have come to the following conclusions according to the objective and topic of this study: 144 • • • • With the non-simultaneous distance instruction system, the researcher integrated affective computing with the instruction of a video course, and developed the affective learning system. The usability of the system scored 73.75, meaning that average users were satisfied with the usability of the system. Therefore, the researcher believes that the usability of the non-simultaneous distance instruction system with affective computing was high. This study used triangulation to investigate respondents’ satisfaction with the affective learning system. In the interviews, most respondents said that the system was simple and easy to operate, the interactive agent mechanism made learning interesting, and enhanced their passion for learning. Hence, the researcher thinks that most of the respondents were satisfied with the affective learning system. According to the results of affective computing, the researcher collected the negative emotions of the respondents, and started the training game at an appropriate time in order that the respondents could change their emotion during learning. Most of the respondents believed that such a mechanism was helpful to learning at a later stage, thus, the researcher believes that the training game in this system can effectively improve emotion during learning. This study aims to develop an affective learning system based on a non-simultaneous distance instruction system. To determine if the system could maintain and enhance the learning effectiveness of the existing system, the researcher divided the respondents into two groups, namely, the Experimental group and the Control group, and conducted simultaneous assessment of learning effectiveness. According to the experiment, the affective learning system not only maintained the existing learning effectiveness, but also improved the learning effectiveness of most respondents. Therefore, this research believes that the affective learning system can effectively enhance learning effectiveness. The research results and conclusions show the possibility of the popularization of this affective learning system, and offers suggestions for relevant future research. Suggestions for the affective learning system During the experiments, the researcher found that the response mechanism of the interactive agent was a key factor that could effectively increase the respondents’ use of the system. If the recognition and accuracy of the existing affective module are enhanced, respondents will feel a stronger intention to use the system, and will not feel bored in the later stages of learning. Regarding the Chinese meaning emotion recognition module, the researcher thinks that more daily expressions can be added to strengthen the understanding of the system, and that a diverse response mechanism can be added to lengthen the interaction between respondents and the interactive agent. Meanwhile, a large quantity of daily life facial expressions should be collected in order to enrich the database, which in turn will enhance the accuracy of real-time facial expression recognition. Regarding the physiological signal recognition module, the researchers thinks that, if it is impossible to buy specialized devices, such as an affective lab, the physiological signal recognition module should be removed in order to promote the efficacy of the entire system. Regarding the brainwave concentration and relaxation module, the researcher believes that both concentration and relaxation are important factors that influence learning emotion, and that if the two are incorporated into the realtime detection of the system, it will significantly facilitate learning. With non-simultaneous distance instruction concepts, this study adopted video-audio materials for instruction. It is suggested that future researchers make full use of the functions of the non-simultaneous distance instruction system, and add curricular interaction modules, such as a discussion and curricular interaction sections in order to enhance the interaction between teachers and students, stimulate an actual instruction environment, and strengthen the on-site experience. Suggestions for video-audio course material The video-audio course material in this study was the PPT of the textbook instructed by teachers, but without the reality of face-to-face instruction; hence, the researcher thinks that interactive materials, such as augmented reality, should be added to involve respondents in the instruction and reduce boredom. The researcher believes this affective learning system will attract more users. Additionally, pictures of teacher instruction can be added into the videoaudio course section to enhance the on-site experience. Regarding the assessment of learning effectiveness, the researcher suggests that future researchers follow the mode, as this study involved the demonstrations of experts, which indicate a high-level of reliability and validity. 145 Suggestions for the interactive agent The interactive agent of this study is a singular role, thus, the researcher suggests that a paper doll system be added by future research in order that respondents will have an alternative. Meanwhile, additional accessories for the interactive agent can be added to enhance respondents’ identification with the interactive agent, and thus, improve their learning experience. In terms of the emotion of the interactive agent, more emotional behaviors, such as jumping with pleasure, can be added to reinforce the visual experience. References Ammar, M. B., Neji, M., Alimi, A. M., & Gouardères, G. (2010). The Affective tutoring system. Expert Systems with Applications, 37, 3013-3023. Basu, T., & Murthy, C. A. (2012). Effective text classification by a supervised feature selection approach. In 2012 IEEE 12th International Conference on Data Mining Workshops (pp. 918-925). doi:10.1109/ICDMW.2012.45 Bangor, A., Kortum, P., & Miller, J. A. (2009). Determining what individual SUS scores mean: Adding an adjective rating scale. Journal of Usability Studies, 4(3), 114-123. Chen, Y. S., Kao, T. C., & Sheu, J. P. (2003). A Mobile learning system for scaffolding bird watching learning. Journal of Computer Assisted Learning, 19(3), 347-359. Ekman, P., & Friesen, W. V. (1971). Constants across cultures in the face and emotion. Journal of personality and social psychology, 17(2), 124-129. Eyharabide, V., & Amandi, A. (2012). Ontology-based user profile learning. Applied Intelligence, 36(4), 857-869. Ezhilarasi, R., & Minu, R. I. (2012). Automatic emotion recognition and classification. Procedia Engineering, 38(0), 21-26. Feldman, R. (2013). Techniques and applications for sentiment analysis. Communications of the ACM, 56(4), 82-89. Gunes, H., & Piccardi, M. (2007). Bi-model emotion recognition from expressive face and body gestures. Journal of Network and Computer Applications, 30, 1334-1345. Hsu, K. C., Lin, H. C. K., Lin, I. L., & Lin, J. W. (2014). The Design and evaluation of an affective tutoring system. Journal of Internet Technology, 15(4), 533-542. Islam, A. K. M. N. (2013). Investigating e-learning system usage outcomes in the university context. Computers & Education, 69, 387-399. Infante, C., Weitz, J., Reyes, T., Nussbaum, M., Gómez, F., & Radovic, D. (2009). Co-located collaborative learning video game with single display groupware. Interactive Learning Environments, 18, 177 -195. Kerr, M. S., Rynearson, K., & Kerr, M. C. (2006). Student characteristics for online learning success. The Internet and Higher Education, 9(2), 91-105. Liaw, S. S., Chen, G. D., & Huang, H. M. (2008). Users’ attitudes toward Web-based Collaborative learning systems for knowledge management. Computers & Education, 50, 950-961. Lin, H. C. K., Wang, C., H., Chao, C. J., & Chien, M. K. (2012). Employing textual and facial emotion recognition to design an affective tutoring system. The Turkish Online Journal of Educational Technology, 11(4), 418-426. Lin, H. C. K., Chen, N. S., Sun, R. T., & Tsai, I. H. (2012). Usability of affective interfaces for a digital arts tutoring system. Behaviour & Information Technology, 33(2), 105-106. Lin, H. C. K., Hsieh, M. C., Loh, L. C., & Wang, C. H. (2012). An Emotion recognition mechanism based on the combination of mutual information and semantic clues. Journal of Ambient Intelligence and Humanized Computing, 3(1), 19-29. Lin, H. C. K., Wu, C. H., & Hsueh, Y. P. (2014). The Influence of using affective tutoring system in accounting remedial instruction on learning performance and usability. Computers in Human Behavior, 41, 514-522. Lin, H. C. K., Tsai, S. C., Cheng, Y. C., Chao, C. J., & Su, S. H. (2014). Usability evaluation of affective tutoring systems on web page. Mitteilungen Klosterneuburg, 64(6), 27-40. Liu, B. (2012). Sentiment analysis and opinion mining. Synthesis Lectures on Human Language Technologies, 5(1), 1-167. 146 Lu, C. Y., Lin, S. H., Liu, J. C., Cruz-Lara, S., & Hong, J. S. (2010). Automatic event-level textual emotion sensing using mutual action histogram between entities. Expert Systems with Applications, 37(2), 1643-1653. Mao X., & Li, Z. (2010). Agent based affective tutoring systems: A Pilot study. Computers & Education, 55, 202-208. Martin, F., Pastore, R., & Snider, J. (2012). Developing mobile based instruction. Techtrends: Linking Research and Practice To Improve Learning, 56(5), 46-51. Metri, P., Ghorpade, J., & Butalia, A. (2012). Facial emotion recognition using context based multimodal approach. International Journal of Emerging Sciences, 2(1), 171-182. Nielsen, J. (1994). Heuristic evaluation. In J. Nielsen, & R. L. Mack (Eds.), Usability Inspection Methods. New York, NY: John Wiley & Sons. Peng, H. Y., Chuang, P. Y., Hwang, G. J., Chu, H. C., Wu, T. T., & Huang, S. X. (2009). Ubiquitous performance-support system as Mindtool: A Case study of instructional decision making and learning assistant. Educational Technology & Society, 12(1), 107120. Picard, R. W. (2000). Affective computing. Cambridge, UK: MIT Press. Picard, R. W., & Klein, J. (2002). Computers that recognise and respond to user emotion: Theoretical and practical implications. Interacting with Computers, 14(2), 141-169. Sarrafzadeh,A., Alexander, S., Dadgostar, F., Fan, C., & Bigdeli, A. (2008). “How do you know that I don’t understand?” A Look at the future of intelligent tutoring systems. Computers in Human Behavior, 24, 1342-1363. Yan, J., Bracewell, D. B., Ren, F., & Kuroiwa, S. (2008). The Creation of a Chinese emotion ontology based on HowNet. Engineering Letters, 16(1), 166-171. 147